DE102013209736A1 - Method for evaluating obstacles in a driver assistance system for motor vehicles - Google Patents

Abstract

Method for evaluating obstacles based on location data of a radar sensor (10) in a driver assistance system for motor vehicles, in which at least one evaluation function (R1, R2) is calculated, which relates to a set of measured variables (Xi) relating to a potential obstacle , indicates whether the potential obstacle is to be assessed as a genuine obstacle, characterized in that a complexity indicator (k) indicating the complexity of a current measurement situation is formed from the location data and that at least two different evaluation functions (R1 , R2) and, depending on the complexity indicator (k), it is decided which of the evaluation functions is used in the current measurement situation.

Description

The invention relates to a method for evaluating obstacles based on location data of a radar sensor in a driver assistance system for motor vehicles, in which at least one evaluation function is calculated, which indicates whether the potential obstacle, depending on a set of measures relating to a potential obstacle as a real obstacle.

In driver assistance systems, for example in collision warning systems, a radar sensor is often used for locating objects in the environment, in particular in the run-up to the own vehicle, which, for example, according to the FMCW principle (Frequency Modulated Continuous Wave) in a position, distances and relative velocities of the located To measure objects. It can be seen from the fact that the relative speed measured by the radar sensor is considerably smaller than the intrinsic speed of the own vehicle. If, on the other hand, the measured relative velocity of an object is negative (approximation) and equal in magnitude to the airspeed, then it is a stationary object that represents a potential obstacle. Due to a certain angular resolution, the radar sensor is also able to determine whether the object is on the road or at the edge of the road. However, a stationary object, even if it is located on the road, not necessarily be a real obstacle. For example, even relatively small objects such as on-road cans or manhole covers recessed in the pavement cause a radar echo that is difficult to distinguish from the radar echo of a broader object representing a true obstacle.

In order nevertheless to allow a distinction between real obstacles and apparent obstacles, one or more evaluation functions are used in known driver assistance systems, the arguments of which are formed by measured variables which typically have different values for real obstacles than for apparent obstacles.

An example of such a measure is about the signal strength of a single located object. The idea behind this is that an extended object, which is more of a real obstacle, will generally produce a stronger radar echo than a small-scale object, such as a small on-track object that can easily be overrun. The score function may then either take the value 1, depending on the signal strength of the object, which means that the object is considered to be a real obstacle, or the value 0, which means that the object is judged to be a dummy obstacle. However, depending on the embodiment, the weighting function may also assume intermediate values between 0 and 1 that indicate different probabilities of the object being a real obstacle. In order to ultimately reach a clear decision, several evaluation functions based on different criteria are generally linked.

Other examples of measures that say something about the relevance of an object as an obstacle are the elevation angle at which the object is seen by the radar sensor and temporal changes of this elevation angle. A real obstacle will generally be at least level with the radar sensor, so the elevation angle will be zero or positive, while a small object lying on the road will generally have a negative elevation angle, which will increase as the approach approaches Object moved even further to the negative side. On the other hand, with a relatively large real obstacle, it will often happen that the radar echo is received from different elevation angles due to multipath propagation across the ground. Furthermore, pitching and steering motions of one's own vehicle cause the signal to be received by different reflection targets, which also have different elevation angles. A strongly fluctuating elevation angle therefore indicates a real obstacle.

Even if several criteria, ie several evaluation functions, are combined, in practice it will not be possible to correctly evaluate all occurring objects. The evaluation functions must therefore be defined in such a way that a meaningful compromise is found between a high hit rate, ie a high proportion of real obstacles that are actually recognized as obstacles, and a low false warning rate, ie the lowest possible proportion of false objects. which are falsely judged to be real obstacles. For this purpose, parameters which determine the properties of the evaluation function must be suitably selected. In the simplest case, the weighting function is a threshold function that jumps from 0 to 1 when a certain threshold is exceeded. In that case, the threshold represents a parameter that must be chosen appropriately. For more complicated evaluation functions, the parameters may vary to trade sets of multiple thresholds or polynomial coefficients or the like.

In the definition of these parameters, the prevention of false alarms is generally the highest priority, as frequent false warnings significantly reduce the acceptance of the system and even driver assistance systems that actively intervene in the vehicle's braking system in acute collision hazard, even a source of danger. However, if, in order to avoid such false warnings, the parameters are chosen to be "conservative", inevitably a higher proportion of real obstacles will not be detected, so the utility of the system will decrease.

The object of the invention is to specify a method which allows an increase in the usefulness of the driver assistance system with a low frequency of false alarms.

This object is achieved according to the invention in that based on the location data a complexity indicator is formed, which specifies the complexity of a current measurement situation, and at least two different evaluation functions are defined for the same set of measured variables and it is decided in dependence on the complexity indicator which of these evaluation functions the current measurement situation is applied.

The invention is based on the consideration that the totality of the signals that a radar sensor receives at a given time can differ considerably in its complexity from one situation to another. In the simplest cases, only a single radar echo is received by an approximately point-like object, the signal is less noisy and corresponds to sharply defined values not only for the distance and the relative velocity but also for the azimuth angle and elevation angle. In that case, a relatively reliable distinction between true obstacles and apparent obstacles is possible, so that the risk of false warnings is low and it would be useful to parameterize the evaluation functions with a view to maximum benefit. In more complex situations, for example, if several objects are located at the same time, whose radar echoes partially overlap or interfere with each other, the probability of error in the assessment is much greater, so that it would be useful to configure the parameters with regard to avoiding false warnings and for that to take greater benefit losses into account. According to the invention, the parameterization is not fixed immutably for all possible measurement situations, but varies depending on the situation, especially depending on the complexity. This makes it possible, at least in situations with low complexity, to achieve a greater benefit, which then statistically also improves the overall benefit of averaging over a variety of different complex situations.

Advantageous embodiments and further developments of the invention are specified in the subclaims.

In the following embodiments are explained in detail with reference to the drawing.

Show it:

1 a block diagram of a driver assistance system in which the inventive method is implemented;

2 and 3 Example of different parameterizations of a weighting function;

4 a ROC (receiver operating characteristic) diagram for the evaluation functions 2 and 3 ;

5 and 6 ROC diagrams for measurement situations with different complexity;

7 a comparison of ROC diagrams for situation-independent and situation-dependent parameterizations of a weighting function;

8th an example of a spectrum of an intermediate frequency signal in an FMCW radar;

9 a diagram for explaining the angle measurement in an angle-resolving radar sensor;

10 a diagram illustrating an obstacle assessment procedure using several evaluation functions; and

11 a diagram illustrating a procedure for generating a hierarchy of complexity indicators.

At the in 1 The driver assistance system illustrated as a block diagram is, for example, a collision warning system. Essential components of this system are a radar sensor 10 (FMCW radar), a preprocessing stage 12 and an assessment module 14 , In the preprocessing stage 12 become the ones from the radar sensor 10 delivered signals evaluated, and it is formed a number of measures Xi, which is a current measurement situation of the radar sensor 10 characterize. Typically, the measures include Xi as well Location data of one or more objects currently being monitored by the radar sensor 10 were located.

In the assessment module 14 For example, the measured quantities Xi are further evaluated in order to evaluate the located objects (at least the stationary objects) as obstacles in their relevance, so that it can be decided whether a given object represents a real obstacle or not.

Other components of the collision warning system, which are not shown here, are then used to calculate the probability of collision on the basis of the location data for the obstacles assessed as genuine and based on the dynamic data of the own vehicle and if necessary issue an acoustic or haptic collision warning to the driver and / or actively intervene in the braking system of the vehicle and trigger emergency braking.

The radar sensor 10 in the example shown has two transmit antennas 16 . 18 and four receiving antennas 20 on. The receiving antennas are arranged at certain intervals on a horizontal line, so that the azimuth angle of a located radar target can be determined by comparing the amplitudes and phases of the signals received by the various antennas. The transmitting antennas 16 . 18 have an elevation dependency and are via a switching network 22 with a local oscillator 24 connected, so by switching between the transmit antennas 16 and 18 and comparing the received signals for the different switching states and the elevation angle of the located object can be determined at least roughly. The transmitting and receiving antennas 16 . 18 . 20 can be arranged approximately in the focal plane of a beam-forming radar lens, so that a certain directional selectivity is achieved by the more or less large offset of the antennas with respect to the optical axis of the radar lens. However, embodiments are also conceivable in which the transmitting and receiving antennas are designed as planar array antennas. Likewise, embodiments with monostatic antenna concepts are also conceivable.

According to the operating principle of an FMCW radar, the frequency of the local oscillator 24 supplied high-frequency signal ramped modulated. The of the four receiving antennas 20 received radar returns are in four parallel channels in mixers 26 with the oscillator 24 mixed transmit signal so that one receives four intermediate frequency signals S _1-4 , the frequency equal to the difference between the frequency of the received radar echo and the frequency of the radar signal sent at the time of reception. The frequency of the intermediate frequency signal is thus dependent on the one hand on the signal transit time from the transmitting antenna to the object and back to the receiving antennas and on the slope of the modulation ramp and on the other - due to the Doppler effect - also on the relative speed of the located object.

Due to the different path lengths from the object to the individual receiving antennas 20 and optionally the directional selectivity of the receive antennas, the intermediate frequency signals received in the four channels also also have characteristic differences in their (complex) amplitude, which are dependent on the azimuth angle of the object.

In the preprocessing stage 12 The intermediate frequency signals S _{1-4 are} first in an analog / digital converter stage 28 digitized and recorded channel by channel over the duration of a frequency modulation ramp. The time signals thus obtained are then in a transformation stage 30 converted by fast Fourier transformation into corresponding spectra. In each of these spectra, a located object will peak in the form of a peak at a frequency that depends on the distance and the relative velocity of that object. From the frequencies of two peaks belonging to the same object, but obtained on frequency ramps with different pitch, such as a rising ramp and a falling ramp, is then in a downstream evaluation module 32 the distance D and the relative velocity V of the object determined.

If more than two objects are located at the same time, however, a certain ambiguity remains, since it is not immediately clear which peak belongs to which object. However, this ambiguity can be eliminated by considering at least one other spectrum obtained on another modulation ramp with a different ramp slope. The frequency position of each individual peak indicates a linear relationship between the distance and the relative speed of the object in question. This relationship can be represented in a D-V space as a straight line. For the multiple peaks in the multiple spectra, this results in crowds of straight lines, with the at least three straight lines that belong to the same object but come from different modulation ramps, all intersecting in one point, then the true distance and true relative velocity of that object indicates.

In an angle estimation module 34 the amplitudes (magnitudes and phases) of the intermediate frequency signals received in the four channels become object-wise (ie separately for each peak) compared with one another in order to determine the azimuth angles φ of the located objects.

In the preprocessing stage 12 Also integrated is a control module 36 , on the one hand the oscillator 24 controls and determines the frequency modulation, and secondly the switching network 22 drives to switch between the transmit antennas 16 and 18 switch. For example, a frequency modulation ramp with the transmit antenna 16 and then the same ramp once again with the transmitting antenna 18 , In the angle estimation module 34 The amplitudes obtained in this way are also compared with one another in order to determine the elevation angles θ of the objects.

The quantities D, V, φ, θ obtained in this way for each located object form part of the measured quantities Xi. In addition, in the pre-processing stage 12 a few more measured variables are formed, which further characterize the located objects as well as the entire measuring situation. Examples of such measures will be explained later.

The assessment module 14 contains a complexity module 38 in which on the basis of the measured quantities Xi at least one complexity indicator k is formed, which characterizes the complexity of the current measuring situation.

Furthermore, the assessment module contains 14 a functional block 40 in which at least two evaluation functions R1, R2 are stored, which both depend on the same selection of measured variables Xi and provide different values for the relevance R of an object as an obstacle. In this case, the complexity indicator k determines which of the two functions R1, R2 is applied to the measured variables, which is shown in FIG 1 is symbolized by a switch which switches between the functions R1 and R2. With low complexity the function R1 is selected, with higher complexity the function R2 is selected. These functions are defined such that with statistically distributed values of the measured quantities Xi, the function R1 more frequently yields the result that the obstacle is relevant (real) than the function R2.

In 2 and 3 Two simple examples of evaluation functions R1 and R2 are shown. These two evaluation functions belong to a common class in the sense that they depend on the same independent variable. Here, for example, this independent variable is a measured variable X1, which is a measure of the total power received by a located object. This measurand may be in the preprocessing stage 12 For example, be calculated in the manner described below.

8th shows an example of a recorded on a modulation ramp spectrum, ie, a function that the (complex) amplitude A (shown here only the magnitude of the amplitude) as a function of the frequency f indicates. The spectrum contains several peaks 42 . 44 . 46 . 48 which can be assumed to belong to a respective located object. As an example, here is the peak 42 to be viewed as. The total power received by the corresponding object is then the integral of the amplitude square A ² across the peak. As integration limits, for example, the frequencies f- and f + can be taken, in which the amplitude A has decreased to half the peak value (at f0). The measured variable X1 then corresponds to the in 8th hatched area shown.

Since the received power is inversely proportional to the fourth power D ^{4 of} the object, it will generally be convenient to normalize the measurand X 1 to a standard distance by multiplying the value of the integral by a factor equal to the fourth power D of this Object measured distance D ^{4 is} proportional.

A stationary object, which is a real obstacle, for example a vehicle standing on the road, will lead to a relatively pronounced peak, so that the measured variable X1 is correspondingly large. A non-relevant apparent obstacle, such as a manhole cover in the carriageway, on the other hand, will lead to a weak peak, such as the peak 48 in 8th , with a correspondingly low value of the measurand X1. A decision criterion as to whether the object is a real obstacle can therefore be that it checks whether the total power for that object is above or below a certain threshold.

Accordingly, the evaluation function R1 is in 2 a step-like function that has the value 0 for low values of X1 and increases suddenly to the value 1 at a threshold T1 = 0.4. The function value R1 = 1 then means that the associated object is regarded as a real obstacle.

In the 3 The evaluation function R2 shown differs from the evaluation function R1 only in that another threshold T2 = 0.8 is used.

If the evaluation function R1 is applied to the measured variable X1, statistically speaking, a relatively large proportion of the total located stationary objects is assessed as a real obstacle all those for which the quantity X1 is at least 0.4. However, this involves the danger that false warnings often occur, because even an object that is irrelevant in reality is wrongly regarded as a real obstacle.

If, on the other hand, the evaluation function R2 is applied, this risk is less, since only the objects are judged to be real obstacles for which X1 is at least 0.8. In this case, however, there is an increased risk that even real obstacles will not be rated as real (R2 = 1) because their overall performance is still below 0.8. In this case, an actually attached collision warning to the driver would therefore be omitted, so that the benefit of the system would be impaired.

For a class of functions that are dependent on the same independent variables and have the same functional specification, only with different parameters (in this example the thresholds T1, T2) 4 a so-called ROC curve 50 (Receiver Operating Characteristic). This curve establishes, for different values of the parameter, a relationship between the relative frequency Nf of the false warnings and the relative frequency Nw of the correct hits, ie the number of cases in which an object is correctly recognized as a real obstacle. This curve can be empirically determined for a given radar sensor and a given class of evaluation functions R1, R2 by calculating the evaluation functions for a variety of measurement situations covering the full range of measurement situations encountered in daily traffic, with as much reality as possible similar frequency distribution. For example, for this purpose, measuring runs can be carried out in which the vehicle travels a certain amount of time on a highway, a certain amount of time on a country road, some in city traffic, occasionally also in a tunnel, etc., the lengths of the respective times being approximately behave as in a typical "life story" of a motor vehicle. Depending on the choice of the parameter, here the threshold value T1 or T2, one then obtains another value pair (Nf, Nw). In 4 the parameters (the thresholds 0.8, 0.6, ...) are given along the curve. The higher the threshold, the smaller the frequency Nf of the false alarms, but the smaller the number Nw of the true hits. As the thresholds decrease, the frequency Nf of the false alarms increases while the frequency Nw of the true hit approaches the maximum value of 100%.

If the ROC curve is taken for a given class of evaluation functions for the entirety of all possible measurement situations, then the only design freedom in designing the system is that it makes sense for the determining parameter of the functional class, in this example is chosen for the threshold, where between the goals "Nf smallest possible" and "Nw as large as possible" a compromise must be found.

5 and 6 show examples of ROC curves 52 . 54 that were recorded for different selections of measurement situations. In 5 only measurement situations were considered in which the traffic environment is relatively clear and clearly structured, ie, the number of simultaneously located objects is low and the signals received by each object are less noisy and relatively unique, so that a clear determination of the corresponding location data (distance , Relative velocity, azimuth angle, elevation angle) with relatively narrow error tolerances is possible. In this case, the number Nf of the false warnings is low, even if relatively low threshold values are selected, so that for a still considered acceptable frequency Nfmax of false warnings a relatively high number Nw1 of correct hits and thus a correspondingly high benefit is achieved.

In 6 On the other hand, only measurement situations were taken into account in which the traffic environment was relatively unclear (many objects simultaneously located, signals very noisy and disturbed). In these situations, of course, the error rate is greater, so the ROC curve 54 is flatter and at the threshold 0.8, which provides the same frequency Nfmax of false warnings, only a significantly lower frequency Nw2 is achieved on real hits.

The basic idea of the invention consists in the fact that even in actual use of the driver assistance system a distinction is made between measurement situations with low complexity and those with high complexity and, depending on the complexity, different threshold values or, in the general case, different sets of determining parameters for the evaluation functions.

In 7 is an ROC curve 56 This is obtained by differentiating between different degrees of complexity in the choice of the threshold value, for example, for the measurement situations that occur in 5 the low threshold 0.4 selects for the measurement situations in 6 the higher threshold 0.8. For comparison shows 7 also an ROC curve 58 in the event that one does not differentiate between the degrees of complexity and selects a uniform threshold value for all measurement situations, which gives, averaged over all measurement situations, the same relative frequency Nfmax of false warnings. One can see that in this case only a lower relative frequency of correct hits is achieved and thus by the invention (ROC curve 56 ) the utility of the system is increased.

In practice, one usually uses not only a single class of evaluation functions for the valuation of the standing objects, but several classes which depend on different sets of independent variables, and then logically or arithmetically link the results of the individual evaluation functions.

For example, a class of evaluation functions may be defined in which the set of independent variables includes only a single measure X2 indicating the elevation angle measured for the object in question. The evaluation functions can then be similar to 2 and 3 in a threshold comparison, each with different thresholds. Objects whose elevation angle is positive or only slightly negative will then be judged to be true obstacles, while objects whose elevation angle is negative and magnitude large qualify as apparent obstacles, since these objects do not tower high above the road surface and therefore from their own Vehicle can be driven over.

Further examples of evaluation functions would be functions which evaluate changes in time, for example fluctuations, in the elevation angle and / or certain tendencies in the change of the elevation angle as a function of time or object distance.

The evaluation functions in all these cases can also be multi-valued functions, which not only vary between 0 and 1, but can also assume intermediate values. When working with multiple classes of evaluation functions, the final decision as to whether the object is relevant or not can then be made on the basis of a weighted sum or some other function of the results of the individual evaluation functions. The individual evaluation functions can be, for example, multi-valued functions with several levels and correspondingly several threshold values as parameters. In another embodiment, the evaluation functions may also be continuous functions of one or more measured variables, for example polynomials, in which case the polynomial coefficients are parameters which distinguish the individual evaluation functions of the same class and are varied depending on the complexity. In general, the parameters will be chosen depending on the complexity so that, statistically speaking, the number of cases in which an object is assessed as a real obstacle is greater at low complexity than at high complexity.

The following are some examples of suitable definitions of complexity indicators. Basically, it is possible to differentiate between global complexity indicators, which refer to the measurement situation as a whole, and local complexity indicators, which refer to a single located object or a group of located objects.

An example of a global complexity indicator is the total number of objects located at a given time. For example, in the in 8th spectrum shown four peaks 42 . 44 . 46 . 48 recognizable, so that the total number of located objects would cheat four. Since in practice several modulation ramps are traversed in each measuring cycle, in each cycle several spectra are obtained in which the number of located objects can be different due to disturbances. It is therefore expedient to add the peaks of all the spectra recorded in a measuring cycle and to normalize them in relation to a sufficiently high set maximum value, so that a complexity indicator k is obtained which varies between 0 and 1.

Another example of a global complexity indicator is the angular quality averaged over all located objects. In this context is in 9 illustrates the principle of estimating the azimuth angle. Shown are four curves 60 . 62 . 64 . 66 , one for each of the four receiving antennas 20 , which for a standard object located with a standard signal strength at an azimuth angle φ, the dependence of the amplitude A (shown again only the magnitude of the amplitude, but it is also possible a phase-based angle estimation method) of the azimuth angle. Due to the different directional selectivity of the receiving antennas, the curves have their maximum at a different azimuth angle. The amplitudes measured in the four receiving channels for a given object (frequency at the apex of the peak) are in 9 through circles 70 . 72 . 74 . 76 symbolizes, with the circle 70 belongs to the same receiving channel as the curve 60 , The circle 72 to the same channel as the curve 62 , etc .. The circles 70 - 76 are shown on a vertical line at the angular position, where the best match between the measured amplitudes (symbolized by the circles) and the due to the curves 60 - 66 expected amplitudes exists. In the example shown, the best match is at azimuth angle φ = 7.5 °, which thus represents the angle estimate for the object in question. Due to unavoidable disturbances, the correspondence between the measured values and the theoretical values are not exact. The deviation between the measured values and the values indicated by the curves (for example the mean square deviation) is a measure of the angular quality, ie the quality of the angle measurement. The higher the angular quality for a given object, the more reliable is the angle estimate. At the same time, however, the angular quality is also a measure of how strongly the radar echo is noisy or disturbed by the relevant object.

By averaging the angular qualities over all located objects, one obtains a global indicator which can be regarded as a measure of the general quality and reliability of the location data at the given time.

It is convenient to construct a global complexity indicator that is a weighted sum of an object number dependent complexity indicator k _a and a global angle goodness dependent complexity indicator k _w . k = a * k _a + b * k _w where k _a and k _{w can} each vary in the interval [0, 1] and the positive coefficients a and b are selected such that the sum also varies in the interval [0, 1].

Unlike global complexity counters, local complexity counters refer only to a single object or group of nearby objects. An example of a local complexity indicator is about the width of the peak in the spectrum associated with the object in question or the peaks in the multiple spectra obtained on different ramps in the same measurement cycle. As a measure of the width of the peak could be about the distance between f and f + in 8th are taken, preferably normalized to the total power of the peak. While a narrow peak indicates a noisy object with a well-defined distance and well-defined relative velocity, a broader peak may be due to the signal being more noisy or disturbed, or to peaks being superimposed by multiple reflection centers. A wider peak is therefore assigned a higher complexity indicator.

Methods are also known which specifically aim to detect peak overlays in the spectrum. Examples are in EP 2 182 380 B1 described. If such a peak overlay is detected for the peak associated with the object under consideration, the object is assigned a high local complexity indicator.

Another example of a local complexity indicator is the peak density in the spectrum, i.e., the number of peaks found in a given frequency interval in the vicinity of the peak associated with the subject object. In this case, a higher peak density also means greater complexity. A related complexity indicator is the location density in the DV space, for example the number of located objects whose distance D deviates less than, for example, 5 m from the distance measured for the object under consideration and whose relative velocity V is less than, for example, 1 m / s Relative speed of the object is different. In contrast to the peak density discussed above, only those locations are considered that can be unambiguously assigned to a real object, that is to say locations in which the straight lines in the DV diagram essentially intersect at one point.

Another set of local complexity indicators relates to the quality of azimuth and / or elevation angle measurement. There is a certain correlation between peak overlays and the quality of the angle measurement in the sense that when peak overlaps are detected, the angle measurements are generally useless. Therefore, the complexity indicators relating to the angle measurement should preferably be used only when no peak overlay has been detected.

If for the object in question in the determination of the azimuth angle after the in 9 A high degree of accuracy is detected in the illustrated methods, the object is assigned a low complexity indicator, and as the quality of the angle becomes increasingly poor, the complexity indicator becomes larger. Accordingly, a complexity indicator can be formed, which is dependent on the quality of the measurement of the elevation angle.

Since two ramps are used to measure the elevation angle, once with the transmit antenna 16 and once with the transmitting antenna 18 ( 1 ), it is also possible to carry out the measurement of the azimuth angle on both ramps so that two azimuth angle measurements are obtained for the same object, each with its own angular quality. In that case, one can combine the two angular qualities for the azimuth together to form the complexity indicator and additionally combine these with the quality of the measurement of the elevation angle. In addition, the discrepancy between the two measurements of the azimuth angle (with the antenna 16 and the antenna 18 ) lead to an additional increase in the complexity indicator.

Finally, the global and local complexity indicators can be combined into an overall complexity indicator, which then determines the selection of parameters for the evaluation functions. However, it is also possible for each class of evaluation functions to define its own complexity indicator, which then depends primarily on the variables that are also the independent variables in the valuation functions classes.

In 10 three groups of evaluation functions R _k1 (X1), R _k2 (X2) and R _k3 (X3) are shown, each having a different measurand X1, X2 or X3 as an independent variable (where the symbols X1, X2 and X3 can each stand for a set of several variables). Each class of evaluation functions has several functions that differ in the choice of parameters. This choice of parameters is dependent on the respective complexity indicator k1, k2 and k3, respectively, and for each class of evaluation functions a differently defined complexity indicator can be used. For example, the symbol R _{k1 is} in 10 for different weighting functions defined for each value of the complexity indicator k1 by a different set of parameters.

Thus, in this example, one first obtains three different results which indicate, whether in the form of a yes / no statement or in the form of a probability, whether the object in question is a real obstacle or not. To make a final decision (R _k1, R _k2, R _k3) is then a function R calculated, the values of the functions R _k1 (X1), R _k2 (X2) and R _k3 (X3) is dependent and either has the value 1 (relevant obstacle) or the value 0 (no relevant obstacle).

The complexity indicators can also be combined in a hierarchical structure to form an overall complexity indicator, either for all evaluation functions in common or only for a single evaluation function at a time. Illustrated as an example 11 the formation of a total complexity indicator k1 for the evaluation functions R _k1 in 10 , In this example, five intermediate indicators Z1-Z5 are first formed, each of up to five independent variables X _k 1, X _k 2, X _k 3, X _k 4, X _k may depend 5, identified by the location data in the respective measuring cycle (The index k merely symbolizes here that the variables in question serve to calculate complexity indicators and not to calculate evaluation functions.) These variables X _k 1 -X _k 5 can be a selection of the measured quantities (Xi) have been named above in the context of the various examples of complexity indicators, for example the number of objects (for a global complexity indicator), the total angular quality averaged over all objects, and the variables that flow into the local complexity indicators, eg the peak density, Peak width, angular quality in azimuth, angular quality in elevation and discrepancy between the two azimuth angle measurement Each of the intermediate indicators Z1-Z5 depends on at least one of these variables, but may also depend on several of these variables. Each of the intermediate indicators Z1-Z5 can be a weighted sum or a polynomial of the variables X _k i (i = 1-5) and be defined such that the values of the intermediate indicator lie in the interval [0, 1]. From the intermediate indicators Z1-Z5 thus obtained, the final complexity indicator k1 is then formed by a further functional rule, for example for a weighted sum or for a polynomial. Accordingly, the complexity indicators k2 are, _k2, R _k3 k3 formed for the evaluation functions R, each with different operating rules and other selections of independent variables.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• EP 2182380 B1 [0060]

Claims (12)

Method for evaluating obstacles based on location data of a radar sensor ( 10 ) In a driver assistance system for motor vehicles, in which at least one evaluation function (R1, R2; R is calculated _k1, R _k2, R _k3), the dependent (from a set of measured variables X1, X2, X3), based on a potential obstacle indicates whether the potential obstacle is to be regarded as a real obstacle, characterized in that a complexity indicator (k; k1, k2, k3) is formed from the location data, indicating the complexity of a current measurement situation, and that for the same set of measured values (X1; X2; X3) at least two different evaluation functions (R1, R2; R _k1, R _k2, R _k3) are defined and it is decided depending on the complexity indicator, which is applied to the evaluation functions in the current measurement situation. Method according to Claim 1, in which a plurality of classes of evaluation functions (R _k1 , R _k2 , R _k3 ) are defined, from which one evaluation function is selected in each measurement situation depending on the value of the complexity indicator, and the results of the selected evaluation functions then become one Overall result can be combined. Method according to Claim 2, in which a separate complexity indicator (k1, k2, k3) is defined for each class of evaluation functions. Method according to one of the preceding claims, in which several evaluation functions depend on a measured variable (X1) which indicates the signal strength of the radar echo received by a located object. Method according to one of the preceding claims, in which a plurality of evaluation functions e are dependent on a measured quantity (X2) which indicates an elevation angle of the located object. Method according to one of the preceding claims, in which several evaluation functions depend on a measured variable (X3) which indicates a change in the elevation angle measured for the object. Method according to one of the preceding claims, in which a global complexity indicator (k) is formed, which depends on the number of objects located in the current measurement situation and / or on the average quality of the angle measurements for all objects located in this measurement situation. Method according to one of the preceding claims, wherein at least one local complexity indicator is formed, which relates to a single located object and is dependent on an object density in the environment of this object. Method according to one of the preceding claims, for an FMCW radar sensor, in which at least one local complexity indicator is formed which relates to a single located object and indicates whether a superposition of radar echoes is detectable at the location associated with this object in the frequency spectrum of the received signal that come from different objects. Method according to one of the preceding claims, for an angle-resolving radar sensor, in which at least one local complexity indicator is formed, which refers to a single located object and indicates the quality of an angle measurement for this object. Method according to one of the preceding claims, in which at least one complexity indicator (k1) is formed, which is dependent on several intermediate indicators (Z1-Z5), which in turn differ in each case from a selection of variables derived from the location data (X _k 1, X _k 2, X _k 3, X _k 4, X _k 5) are dependent. Driver assistance system for motor vehicles, in which a method according to any one of the preceding claims is implemented.

Images